RDMA first rc pull for v6.7

Primarily rtrs and irdma fixes:
 
 - Fix uninitialized value in ib_get_eth_speed()
 
 - Fix hns refusing to work if userspace doesn't select the correct
   congestion control algorithm
 
 - Several irdma fixes - unreliable Send Queue Drain, use after free, 64k
   page size bugs, device removal races
 
 - Several rtrs bug fixes - crashes, memory leaks, use after free, bad
   credit accounting, bogus WARN_ON
 
 - Typos and a MAINTAINER update
 -----BEGIN PGP SIGNATURE-----
 
 iHUEABYIAB0WIQRRRCHOFoQz/8F5bUaFwuHvBreFYQUCZXMjNwAKCRCFwuHvBreF
 YdnyAP9h6dQG+KgIl26q7Bk1AmsmoseZ1HmeGRacdHsW0qgXPwD/XRX/IgNpl465
 fdt6Qh877LBEjlMFvyMxFTWG7v1ixwY=
 =Tn3a
 -----END PGP SIGNATURE-----

Merge tag 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/rdma/rdma

Pull rdma fixes from Jason Gunthorpe:
 "Primarily rtrs and irdma fixes:

   - Fix uninitialized value in ib_get_eth_speed()

   - Fix hns refusing to work if userspace doesn't select the correct
     congestion control algorithm

   - Several irdma fixes - unreliable Send Queue Drain, use after free,
     64k page size bugs, device removal races

   - Several rtrs bug fixes - crashes, memory leaks, use after free, bad
     credit accounting, bogus WARN_ON

   - Typos and a MAINTAINER update"

* tag 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/rdma/rdma:
  RDMA/irdma: Avoid free the non-cqp_request scratch
  RDMA/irdma: Fix support for 64k pages
  RDMA/irdma: Ensure iWarp QP queue memory is OS paged aligned
  RDMA/core: Fix umem iterator when PAGE_SIZE is greater then HCA pgsz
  RDMA/irdma: Fix UAF in irdma_sc_ccq_get_cqe_info()
  RDMA/bnxt_re: Correct module description string
  RDMA/rtrs-clt: Remove the warnings for req in_use check
  RDMA/rtrs-clt: Fix the max_send_wr setting
  RDMA/rtrs-srv: Destroy path files after making sure no IOs in-flight
  RDMA/rtrs-srv: Free srv_mr iu only when always_invalidate is true
  RDMA/rtrs-srv: Check return values while processing info request
  RDMA/rtrs-clt: Start hb after path_up
  RDMA/rtrs-srv: Do not unconditionally enable irq
  MAINTAINERS: Add Chengchang Tang as Hisilicon RoCE maintainer
  RDMA/irdma: Add wait for suspend on SQD
  RDMA/irdma: Do not modify to SQD on error
  RDMA/hns: Fix unnecessary err return when using invalid congest control algorithm
  RDMA/core: Fix uninit-value access in ib_get_eth_speed()

MAINTAINERS

@@ -9573,6 +9573,7 @@ F:	drivers/crypto/hisilicon/sgl.c
 F:	include/linux/hisi_acc_qm.h
 
 HISILICON ROCE DRIVER
+M:	Chengchang Tang <tangchengchang@huawei.com>
 M:	Junxian Huang <huangjunxian6@hisilicon.com>
 L:	linux-rdma@vger.kernel.org
 S:	Maintained

drivers/infiniband/core/umem.c

@@ -96,12 +96,6 @@ unsigned long ib_umem_find_best_pgsz(struct ib_umem *umem,
 		return page_size;
 	}
 
-	/* rdma_for_each_block() has a bug if the page size is smaller than the
-	 * page size used to build the umem. For now prevent smaller page sizes
-	 * from being returned.
-	 */
-	pgsz_bitmap &= GENMASK(BITS_PER_LONG - 1, PAGE_SHIFT);
-
 	/* The best result is the smallest page size that results in the minimum
 	 * number of required pages. Compute the largest page size that could
 	 * work based on VA address bits that don't change.

drivers/infiniband/core/verbs.c

@@ -1971,7 +1971,7 @@ int ib_get_eth_speed(struct ib_device *dev, u32 port_num, u16 *speed, u8 *width)
 	int rc;
 	u32 netdev_speed;
 	struct net_device *netdev;
-	struct ethtool_link_ksettings lksettings;
+	struct ethtool_link_ksettings lksettings = {};
 
 	if (rdma_port_get_link_layer(dev, port_num) != IB_LINK_LAYER_ETHERNET)
 		return -EINVAL;

drivers/infiniband/hw/bnxt_re/main.c

@@ -71,7 +71,7 @@ static char version[] =
 		BNXT_RE_DESC "\n";
 
 MODULE_AUTHOR("Eddie Wai <eddie.wai@broadcom.com>");
-MODULE_DESCRIPTION(BNXT_RE_DESC " Driver");
+MODULE_DESCRIPTION(BNXT_RE_DESC);
 MODULE_LICENSE("Dual BSD/GPL");
 
 /* globals */

drivers/infiniband/hw/hns/hns_roce_hw_v2.c

@@ -4760,10 +4760,15 @@ static int check_cong_type(struct ib_qp *ibqp,
 		cong_alg->wnd_mode_sel = WND_LIMIT;
 		break;
 	default:
-		ibdev_err(&hr_dev->ib_dev,
+		ibdev_warn(&hr_dev->ib_dev,
-			  "error type(%u) for congestion selection.\n",
+			   "invalid type(%u) for congestion selection.\n",
-			  hr_dev->caps.cong_type);
+			   hr_dev->caps.cong_type);
-		return -EINVAL;
+		hr_dev->caps.cong_type = CONG_TYPE_DCQCN;
+		cong_alg->alg_sel = CONG_DCQCN;
+		cong_alg->alg_sub_sel = UNSUPPORT_CONG_LEVEL;
+		cong_alg->dip_vld = DIP_INVALID;
+		cong_alg->wnd_mode_sel = WND_LIMIT;
+		break;
 	}
 
 	return 0;

drivers/infiniband/hw/irdma/hw.c

@@ -321,7 +321,11 @@ static void irdma_process_aeq(struct irdma_pci_f *rf)
 			break;
 		case IRDMA_AE_QP_SUSPEND_COMPLETE:
 			if (iwqp->iwdev->vsi.tc_change_pending) {
-				atomic_dec(&iwqp->sc_qp.vsi->qp_suspend_reqs);
+				if (!atomic_dec_return(&qp->vsi->qp_suspend_reqs))
+					wake_up(&iwqp->iwdev->suspend_wq);
+			}
+			if (iwqp->suspend_pending) {
+				iwqp->suspend_pending = false;
 				wake_up(&iwqp->iwdev->suspend_wq);
 			}
 			break;
@@ -581,9 +585,6 @@ static void irdma_destroy_cqp(struct irdma_pci_f *rf)
 	struct irdma_cqp *cqp = &rf->cqp;
 	int status = 0;
 
-	if (rf->cqp_cmpl_wq)
-		destroy_workqueue(rf->cqp_cmpl_wq);
-
 	status = irdma_sc_cqp_destroy(dev->cqp);
 	if (status)
 		ibdev_dbg(to_ibdev(dev), "ERR: Destroy CQP failed %d\n", status);
@@ -748,6 +749,9 @@ static void irdma_destroy_ccq(struct irdma_pci_f *rf)
 	struct irdma_ccq *ccq = &rf->ccq;
 	int status = 0;
 
+	if (rf->cqp_cmpl_wq)
+		destroy_workqueue(rf->cqp_cmpl_wq);
+
 	if (!rf->reset)
 		status = irdma_sc_ccq_destroy(dev->ccq, 0, true);
 	if (status)
@@ -1180,7 +1184,6 @@ static int irdma_create_ceq(struct irdma_pci_f *rf, struct irdma_ceq *iwceq,
 	int status;
 	struct irdma_ceq_init_info info = {};
 	struct irdma_sc_dev *dev = &rf->sc_dev;
-	u64 scratch;
 	u32 ceq_size;
 
 	info.ceq_id = ceq_id;
@@ -1201,14 +1204,13 @@ static int irdma_create_ceq(struct irdma_pci_f *rf, struct irdma_ceq *iwceq,
 	iwceq->sc_ceq.ceq_id = ceq_id;
 	info.dev = dev;
 	info.vsi = vsi;
-	scratch = (uintptr_t)&rf->cqp.sc_cqp;
 	status = irdma_sc_ceq_init(&iwceq->sc_ceq, &info);
 	if (!status) {
 		if (dev->ceq_valid)
 			status = irdma_cqp_ceq_cmd(&rf->sc_dev, &iwceq->sc_ceq,
 						   IRDMA_OP_CEQ_CREATE);
 		else
-			status = irdma_sc_cceq_create(&iwceq->sc_ceq, scratch);
+			status = irdma_sc_cceq_create(&iwceq->sc_ceq, 0);
 	}
 
 	if (status) {

drivers/infiniband/hw/irdma/main.c

@@ -48,7 +48,7 @@ static void irdma_prep_tc_change(struct irdma_device *iwdev)
 	/* Wait for all qp's to suspend */
 	wait_event_timeout(iwdev->suspend_wq,
 			   !atomic_read(&iwdev->vsi.qp_suspend_reqs),
-			   IRDMA_EVENT_TIMEOUT);
+			   msecs_to_jiffies(IRDMA_EVENT_TIMEOUT_MS));
 	irdma_ws_reset(&iwdev->vsi);
 }
 

drivers/infiniband/hw/irdma/main.h

@@ -78,7 +78,7 @@ extern struct auxiliary_driver i40iw_auxiliary_drv;
 
 #define MAX_DPC_ITERATIONS	128
 
-#define IRDMA_EVENT_TIMEOUT		50000
+#define IRDMA_EVENT_TIMEOUT_MS		5000
 #define IRDMA_VCHNL_EVENT_TIMEOUT	100000
 #define IRDMA_RST_TIMEOUT_HZ		4
 

drivers/infiniband/hw/irdma/verbs.c

@@ -1157,6 +1157,21 @@ exit:
 	return prio;
 }
 
+static int irdma_wait_for_suspend(struct irdma_qp *iwqp)
+{
+	if (!wait_event_timeout(iwqp->iwdev->suspend_wq,
+				!iwqp->suspend_pending,
+				msecs_to_jiffies(IRDMA_EVENT_TIMEOUT_MS))) {
+		iwqp->suspend_pending = false;
+		ibdev_warn(&iwqp->iwdev->ibdev,
+			   "modify_qp timed out waiting for suspend. qp_id = %d, last_ae = 0x%x\n",
+			   iwqp->ibqp.qp_num, iwqp->last_aeq);
+		return -EBUSY;
+	}
+
+	return 0;
+}
+
 /**
  * irdma_modify_qp_roce - modify qp request
  * @ibqp: qp's pointer for modify
@@ -1420,17 +1435,11 @@ int irdma_modify_qp_roce(struct ib_qp *ibqp, struct ib_qp_attr *attr,
 
 			info.next_iwarp_state = IRDMA_QP_STATE_SQD;
 			issue_modify_qp = 1;
+			iwqp->suspend_pending = true;
 			break;
 		case IB_QPS_SQE:
 		case IB_QPS_ERR:
 		case IB_QPS_RESET:
-			if (iwqp->iwarp_state == IRDMA_QP_STATE_RTS) {
-				spin_unlock_irqrestore(&iwqp->lock, flags);
-				info.next_iwarp_state = IRDMA_QP_STATE_SQD;
-				irdma_hw_modify_qp(iwdev, iwqp, &info, true);
-				spin_lock_irqsave(&iwqp->lock, flags);
-			}
-
 			if (iwqp->iwarp_state == IRDMA_QP_STATE_ERROR) {
 				spin_unlock_irqrestore(&iwqp->lock, flags);
 				if (udata && udata->inlen) {
@@ -1467,6 +1476,11 @@ int irdma_modify_qp_roce(struct ib_qp *ibqp, struct ib_qp_attr *attr,
 			ctx_info->rem_endpoint_idx = udp_info->arp_idx;
 			if (irdma_hw_modify_qp(iwdev, iwqp, &info, true))
 				return -EINVAL;
+			if (info.next_iwarp_state == IRDMA_QP_STATE_SQD) {
+				ret = irdma_wait_for_suspend(iwqp);
+				if (ret)
+					return ret;
+			}
 			spin_lock_irqsave(&iwqp->lock, flags);
 			if (iwqp->iwarp_state == info.curr_iwarp_state) {
 				iwqp->iwarp_state = info.next_iwarp_state;
@@ -2900,7 +2914,7 @@ static struct irdma_mr *irdma_alloc_iwmr(struct ib_umem *region,
 	iwmr->type = reg_type;
 
 	pgsz_bitmap = (reg_type == IRDMA_MEMREG_TYPE_MEM) ?
-		iwdev->rf->sc_dev.hw_attrs.page_size_cap : PAGE_SIZE;
+		iwdev->rf->sc_dev.hw_attrs.page_size_cap : SZ_4K;
 
 	iwmr->page_size = ib_umem_find_best_pgsz(region, pgsz_bitmap, virt);
 	if (unlikely(!iwmr->page_size)) {
@@ -2932,6 +2946,11 @@ static int irdma_reg_user_mr_type_qp(struct irdma_mem_reg_req req,
 	int err;
 	u8 lvl;
 
+	/* iWarp: Catch page not starting on OS page boundary */
+	if (!rdma_protocol_roce(&iwdev->ibdev, 1) &&
+	    ib_umem_offset(iwmr->region))
+		return -EINVAL;
+
 	total = req.sq_pages + req.rq_pages + 1;
 	if (total > iwmr->page_cnt)
 		return -EINVAL;

drivers/infiniband/hw/irdma/verbs.h

@@ -198,6 +198,7 @@ struct irdma_qp {
 	u8 flush_issued : 1;
 	u8 sig_all : 1;
 	u8 pau_mode : 1;
+	u8 suspend_pending : 1;
 	u8 rsvd : 1;
 	u8 iwarp_state;
 	u16 term_sq_flush_code;

drivers/infiniband/ulp/rtrs/rtrs-clt.c

@@ -384,7 +384,7 @@ static void complete_rdma_req(struct rtrs_clt_io_req *req, int errno,
 	struct rtrs_clt_path *clt_path;
 	int err;
 
-	if (WARN_ON(!req->in_use))
+	if (!req->in_use)
 		return;
 	if (WARN_ON(!req->con))
 		return;
@@ -1699,7 +1699,7 @@ static int create_con_cq_qp(struct rtrs_clt_con *con)
 		clt_path->s.dev_ref++;
 		max_send_wr = min_t(int, wr_limit,
 			      /* QD * (REQ + RSP + FR REGS or INVS) + drain */
-			      clt_path->queue_depth * 3 + 1);
+			      clt_path->queue_depth * 4 + 1);
 		max_recv_wr = min_t(int, wr_limit,
 			      clt_path->queue_depth * 3 + 1);
 		max_send_sge = 2;
@@ -2350,8 +2350,6 @@ static int init_conns(struct rtrs_clt_path *clt_path)
 	if (err)
 		goto destroy;
 
-	rtrs_start_hb(&clt_path->s);
-
 	return 0;
 
 destroy:
@@ -2625,6 +2623,7 @@ static int init_path(struct rtrs_clt_path *clt_path)
 		goto out;
 	}
 	rtrs_clt_path_up(clt_path);
+	rtrs_start_hb(&clt_path->s);
 out:
 	mutex_unlock(&clt_path->init_mutex);
 

drivers/infiniband/ulp/rtrs/rtrs-srv.c

@@ -65,8 +65,9 @@ static bool rtrs_srv_change_state(struct rtrs_srv_path *srv_path,
 {
 	enum rtrs_srv_state old_state;
 	bool changed = false;
+	unsigned long flags;
 
-	spin_lock_irq(&srv_path->state_lock);
+	spin_lock_irqsave(&srv_path->state_lock, flags);
 	old_state = srv_path->state;
 	switch (new_state) {
 	case RTRS_SRV_CONNECTED:
@@ -87,7 +88,7 @@ static bool rtrs_srv_change_state(struct rtrs_srv_path *srv_path,
 	}
 	if (changed)
 		srv_path->state = new_state;
-	spin_unlock_irq(&srv_path->state_lock);
+	spin_unlock_irqrestore(&srv_path->state_lock, flags);
 
 	return changed;
 }
@@ -550,7 +551,10 @@ static void unmap_cont_bufs(struct rtrs_srv_path *srv_path)
 		struct rtrs_srv_mr *srv_mr;
 
 		srv_mr = &srv_path->mrs[i];
-		rtrs_iu_free(srv_mr->iu, srv_path->s.dev->ib_dev, 1);
+
+		if (always_invalidate)
+			rtrs_iu_free(srv_mr->iu, srv_path->s.dev->ib_dev, 1);
+
 		ib_dereg_mr(srv_mr->mr);
 		ib_dma_unmap_sg(srv_path->s.dev->ib_dev, srv_mr->sgt.sgl,
 				srv_mr->sgt.nents, DMA_BIDIRECTIONAL);
@@ -709,20 +713,23 @@ static void rtrs_srv_info_rsp_done(struct ib_cq *cq, struct ib_wc *wc)
 	WARN_ON(wc->opcode != IB_WC_SEND);
 }
 
-static void rtrs_srv_path_up(struct rtrs_srv_path *srv_path)
+static int rtrs_srv_path_up(struct rtrs_srv_path *srv_path)
 {
 	struct rtrs_srv_sess *srv = srv_path->srv;
 	struct rtrs_srv_ctx *ctx = srv->ctx;
-	int up;
+	int up, ret = 0;
 
 	mutex_lock(&srv->paths_ev_mutex);
 	up = ++srv->paths_up;
 	if (up == 1)
-		ctx->ops.link_ev(srv, RTRS_SRV_LINK_EV_CONNECTED, NULL);
+		ret = ctx->ops.link_ev(srv, RTRS_SRV_LINK_EV_CONNECTED, NULL);
 	mutex_unlock(&srv->paths_ev_mutex);
 
 	/* Mark session as established */
-	srv_path->established = true;
+	if (!ret)
+		srv_path->established = true;
+
+	return ret;
 }
 
 static void rtrs_srv_path_down(struct rtrs_srv_path *srv_path)
@@ -851,7 +858,12 @@ static int process_info_req(struct rtrs_srv_con *con,
 		goto iu_free;
 	kobject_get(&srv_path->kobj);
 	get_device(&srv_path->srv->dev);
-	rtrs_srv_change_state(srv_path, RTRS_SRV_CONNECTED);
+	err = rtrs_srv_change_state(srv_path, RTRS_SRV_CONNECTED);
+	if (!err) {
+		rtrs_err(s, "rtrs_srv_change_state(), err: %d\n", err);
+		goto iu_free;
+	}
+
 	rtrs_srv_start_hb(srv_path);
 
 	/*
@@ -860,7 +872,11 @@ static int process_info_req(struct rtrs_srv_con *con,
 	 * all connections are successfully established.  Thus, simply notify
 	 * listener with a proper event if we are the first path.
 	 */
-	rtrs_srv_path_up(srv_path);
+	err = rtrs_srv_path_up(srv_path);
+	if (err) {
+		rtrs_err(s, "rtrs_srv_path_up(), err: %d\n", err);
+		goto iu_free;
+	}
 
 	ib_dma_sync_single_for_device(srv_path->s.dev->ib_dev,
 				      tx_iu->dma_addr,
@@ -1516,7 +1532,6 @@ static void rtrs_srv_close_work(struct work_struct *work)
 
 	srv_path = container_of(work, typeof(*srv_path), close_work);
 
-	rtrs_srv_destroy_path_files(srv_path);
 	rtrs_srv_stop_hb(srv_path);
 
 	for (i = 0; i < srv_path->s.con_num; i++) {
@@ -1536,6 +1551,8 @@ static void rtrs_srv_close_work(struct work_struct *work)
 	/* Wait for all completion */
 	wait_for_completion(&srv_path->complete_done);
 
+	rtrs_srv_destroy_path_files(srv_path);
+
 	/* Notify upper layer if we are the last path */
 	rtrs_srv_path_down(srv_path);
 

include/rdma/ib_umem.h

@@ -77,6 +77,13 @@ static inline void __rdma_umem_block_iter_start(struct ib_block_iter *biter,
 {
 	__rdma_block_iter_start(biter, umem->sgt_append.sgt.sgl,
 				umem->sgt_append.sgt.nents, pgsz);
+	biter->__sg_advance = ib_umem_offset(umem) & ~(pgsz - 1);
+	biter->__sg_numblocks = ib_umem_num_dma_blocks(umem, pgsz);
+}
+
+static inline bool __rdma_umem_block_iter_next(struct ib_block_iter *biter)
+{
+	return __rdma_block_iter_next(biter) && biter->__sg_numblocks--;
 }
 
 /**
@@ -92,7 +99,7 @@ static inline void __rdma_umem_block_iter_start(struct ib_block_iter *biter,
  */
 #define rdma_umem_for_each_dma_block(umem, biter, pgsz)                        \
 	for (__rdma_umem_block_iter_start(biter, umem, pgsz);                  \
-	     __rdma_block_iter_next(biter);)
+	     __rdma_umem_block_iter_next(biter);)
 
 #ifdef CONFIG_INFINIBAND_USER_MEM
 

include/rdma/ib_verbs.h

@@ -2850,6 +2850,7 @@ struct ib_block_iter {
 	/* internal states */
 	struct scatterlist *__sg;	/* sg holding the current aligned block */
 	dma_addr_t __dma_addr;		/* unaligned DMA address of this block */
+	size_t __sg_numblocks;		/* ib_umem_num_dma_blocks() */
 	unsigned int __sg_nents;	/* number of SG entries */
 	unsigned int __sg_advance;	/* number of bytes to advance in sg in next step */
 	unsigned int __pg_bit;		/* alignment of current block */